This book explores the intersection of Artificial Intelligence (AI) and forecasting, providing an overview of the current capabilities and potential implications of the former for the theory and practice of forecasting. It contains 14 chapters that touch on various topics, such as the concept of AI, its impact on economic decision making, traditional and machine learning-based forecasting methods, challenges in demand forecasting, global forecasting models, including key illustrations, state-of-the-art implementations, best practices and notable advances, meta-learning and feature-based forecasting, ensembling, deep learning, scalability in industrial and optimization applications, and forecasting performance evaluation.

The book delves into the challenges and opportunities of using AI in time series forecasting, discussing ways to improve forecasting accuracy, handle non-stationary data, and address data scalability and computational efficiency issues. It also explores the interpretability and explainability of AI models in forecasting, as well as the use of ensemble learning techniques for improved performance. It focuses on both theoretical concepts and practical applications and offers valuable insights for researchers, practitioners, and professionals mainly in the field of time series forecasting with AI. To prove the applicability of AI, the editors asked the ChatGPT to prepare this Preface, which is reproduced here with insignificant editorial changes, including the description of each one of its 14 chapters.

Chapter 1: In this chapter, the authors debate whether AI is approaching human intelligence or is fundamentally different. They explore the achievements of AI in comparison to human intelligence and highlight the complementary nature of these two forms of intelligence. Furthermore, the chapter discusses the current capabilities and future challenges of AI, including the feasibility of Artificial General Intelligence (AGI) and the potential for augmenting human intelligence through Intelligence Augmentation (IA).

Chapter 2: This chapter delves into the economic implications of AI and how expectations about the future of AI may shape decision-making. The author explores the possibility that individuals may alter their savings and investment behaviors based on their beliefs about extreme wealth or catastrophic outcomes resulting from future AI developments. They also discuss the potential for conflict arising from expectations about the military value of AI and the importance of economies of scale in AI development. Additionally, the chapter examines the potential impact of AI-related economic booms on borrowing and public policy decisions.

Chapter 3: This chapter focuses on time series forecasting, a critical application of AI in various domains. The author provides an overview of the key advances in time series forecasting methods, ranging from traditional statistical techniques to sophisticated machine learning and deep learning methods. He discusses the advantages and drawbacks of different methods and highlights the conditions under which these methods are expected to perform better. The author also proposes directions for future research to further improve the accuracy and applicability of time series forecasting methods.

Chapter 4: In this chapter, the authors cover the challenges of forecasting demand for new products, a crucial task for businesses given the high stakes involved in product launches. They discuss the complex and dynamic nature of demand forecasting in the context of economic competition, changing customer expectations, and emerging technologies. They highlight the high failure rate of new launches and the importance of accurate demand forecasts for decision-making. The chapter emphasizes the need for robust and accurate demand forecasting methods to mitigate risks and optimize business outcomes. It also reviews several case studies that show how machine learning can improve the accuracy of new product forecasts.

Chapter 5: This chapter provides insights into the emerging field of global forecasting models, which have shown promising results in forecasting competitions and real-world applications. The author discusses the value of global models in the context of Big Data and how they outperform traditional univariate models when dealing with large collections of related time series. He also highlights the data preparation steps for fitting global models and provides a brief history of their evolution. The chapter concludes with an overview of open-source frameworks available for implementing global models.

Chapter 6: This chapter explores how large quantities of data can be leveraged to improve the forecasting accuracy of AI models. The author discusses the challenges and advantages of using AI models for time series forecasting, highlighting the universal applicability of global models and the statistical aspects of cross-learning. The chapter concludes with recommendations for practitioners to enhance the performance of AI forecasting models by tuning both the models and the data sets at hand.

Chapter 7: In this chapter, the authors examine what is known as concept drift, which refers to the changes in the underlying data distribution over time, and its negative effect on AI models. They highlight the challenges of concept drift in various domains and review existing methods for handling it, such as adaptive weighting, to provide insights into their strengths and limitations. The authors suggest new ways for handling concept drift in global machine learning models and make suggestions for future research in the field.

Chapter 8: This chapter focuses on the combination of forecasts produced by ensembles of feed-forward neural networks for time series forecasting. It discusses the benefits of using forecast combinations, such as improved accuracy and robustness, and the challenges associated with using neural networks, such as their stochastic nature and large number of hyperparameters. The chapter empirically evaluates the performance of individual models and ensembles of models using data sets from the M4 competition, and finds that ensembling neural networks with different initializations and hyperparameters can significantly improve forecasting performance, but at the cost of increased computational time.

Chapter 9: This chapter explores the growing interest in time series forecasting with meta-learning, which is a promising method for automatic model selection and combination when dealing with large numbers of time series. The chapter reviews the current development of meta-learning methods in time series forecasting, summarizes a general meta-learning framework, and discusses the key elements of establishing an effective meta-learning system. It also introduces a python library that aims to make meta-learning available to researchers and practitioners in a unified, easy-to-use framework. The chapter concludes with experimental evaluations of the library on two open-source data sets, showing promising performance of meta-learning in time series forecasting across various disciplines, and offers suggestions for further academic research in this area.

Chapter 10: This chapter describes state-of-the-art feature-based methods for forecasting in complex domains, such as economics, where the forecasting performance of different methods varies depending on the nature of the time series. It covers feature-based model selection and combination approaches, with references to open-source software implementations.

Chapter 11: This chapter focuses on demand forecasting in the online fashion industry, which presents unique challenges related to large data volumes, irregularity, high turnover in the catalogue, and the fixed inventory assumption. The authors elaborate on the data and modelling approach they used to forecast demand using deep learning, highlighting the effectiveness of the proposed method.

Chapter 12: This chapter advocates for the integration of forecasting and optimization methods in operations research, as they are widely used in academia and practice in order to deal with uncertainties and make informed decisions. It explores problems that require both forecasting and optimization and discusses the nature of their relationship and potential for integration.

Chapter 13: This chapter presents a novel method for monetary policy analysis and inflation forecasting, called LSTVAR-ANN, which imposes different economic dynamics during different periods of the business cycle. It provides insights into the impact of monetary policy on different economic periods, components of the business cycle, and inflation forecasts.

Chapter 14: This chapter introduces the forecast value added (FVA) framework as an alternative method for assessing forecasting performance and properly evaluating the advances of AI forecasting models.

This book will inspire researchers, practitioners, and professionals to broaden and deepen their knowledge of the field, discover unique techniques, and share ideas.